1This directory contains the libffi package, which is not part of GCC but
2shipped with GCC as convenience.
3
4Status
5======
6
7libffi-2.00 has not been released yet! This is a development snapshot!
8
9libffi-1.20 was released on October 5, 1998. Check the libffi web
10page for updates: <URL:http://sources.redhat.com/libffi/>.
11
12
13What is libffi?
14===============
15
16Compilers for high level languages generate code that follow certain
17conventions. These conventions are necessary, in part, for separate
18compilation to work. One such convention is the "calling
19convention". The "calling convention" is essentially a set of
20assumptions made by the compiler about where function arguments will
21be found on entry to a function. A "calling convention" also specifies
22where the return value for a function is found.
23
24Some programs may not know at the time of compilation what arguments
25are to be passed to a function. For instance, an interpreter may be
26told at run-time about the number and types of arguments used to call
27a given function. Libffi can be used in such programs to provide a
28bridge from the interpreter program to compiled code.
29
30The libffi library provides a portable, high level programming
31interface to various calling conventions. This allows a programmer to
32call any function specified by a call interface description at run
33time.
34
35Ffi stands for Foreign Function Interface. A foreign function
36interface is the popular name for the interface that allows code
37written in one language to call code written in another language. The
38libffi library really only provides the lowest, machine dependent
39layer of a fully featured foreign function interface. A layer must
40exist above libffi that handles type conversions for values passed
41between the two languages.
42
43
44Supported Platforms and Prerequisites
45=====================================
46
47Libffi has been ported to:
48
49 SunOS 4.1.3 & Solaris 2.x (SPARC-V8, SPARC-V9)
50
51 Irix 5.3 & 6.2 (System V/o32 & n32)
52
53 Intel x86 - Linux (System V ABI)
54
55 Alpha - Linux and OSF/1
56
57 m68k - Linux (System V ABI)
58
59 PowerPC - Linux (System V ABI, Darwin, AIX)
60
61 ARM - Linux (System V ABI)
62
63Libffi has been tested with the egcs 1.0.2 gcc compiler. Chances are
64that other versions will work. Libffi has also been built and tested
65with the SGI compiler tools.
66
67On PowerPC, the tests failed (see the note below).
68
69You must use GNU make to build libffi. SGI's make will not work.
70Sun's probably won't either.
71
72If you port libffi to another platform, please let me know! I assume
73that some will be easy (x86 NetBSD), and others will be more difficult
74(HP).
75
76
77Installing libffi
78=================
79
80[Note: before actually performing any of these installation steps,
81 you may wish to read the "Platform Specific Notes" below.]
82
83First you must configure the distribution for your particular
84system. Go to the directory you wish to build libffi in and run the
85"configure" program found in the root directory of the libffi source
86distribution.
87
88You may want to tell configure where to install the libffi library and
89header files. To do that, use the --prefix configure switch. Libffi
90will install under /usr/local by default.
91
92If you want to enable extra run-time debugging checks use the the
93--enable-debug configure switch. This is useful when your program dies
94mysteriously while using libffi.
95
96Another useful configure switch is --enable-purify-safety. Using this
97will add some extra code which will suppress certain warnings when you
98are using Purify with libffi. Only use this switch when using
99Purify, as it will slow down the library.
100
101Configure has many other options. Use "configure --help" to see them all.
102
103Once configure has finished, type "make". Note that you must be using
104GNU make. SGI's make will not work. Sun's probably won't either.
105You can ftp GNU make from prep.ai.mit.edu:/pub/gnu.
106
107To ensure that libffi is working as advertised, type "make test".
108
109To install the library and header files, type "make install".
110
111
112Using libffi
113============
114
115 The Basics
116 ----------
117
118Libffi assumes that you have a pointer to the function you wish to
119call and that you know the number and types of arguments to pass it,
120as well as the return type of the function.
121
122The first thing you must do is create an ffi_cif object that matches
123the signature of the function you wish to call. The cif in ffi_cif
124stands for Call InterFace. To prepare a call interface object, use the
125following function:
126
127ffi_status ffi_prep_cif(ffi_cif *cif, ffi_abi abi,
128 unsigned int nargs,
129 ffi_type *rtype, ffi_type **atypes);
130
131 CIF is a pointer to the call interface object you wish
132 to initialize.
133
134 ABI is an enum that specifies the calling convention
135 to use for the call. FFI_DEFAULT_ABI defaults
136 to the system's native calling convention. Other
137 ABI's may be used with care. They are system
138 specific.
139
140 NARGS is the number of arguments this function accepts.
141 libffi does not yet support vararg functions.
142
143 RTYPE is a pointer to an ffi_type structure that represents
144 the return type of the function. Ffi_type objects
145 describe the types of values. libffi provides
146 ffi_type objects for many of the native C types:
147 signed int, unsigned int, signed char, unsigned char,
148 etc. There is also a pointer ffi_type object and
149 a void ffi_type. Use &ffi_type_void for functions that
150 don't return values.
151
152 ATYPES is a vector of ffi_type pointers. ARGS must be NARGS long.
153 If NARGS is 0, this is ignored.
154
155
156ffi_prep_cif will return a status code that you are responsible
157for checking. It will be one of the following:
158
159 FFI_OK - All is good.
160
161 FFI_BAD_TYPEDEF - One of the ffi_type objects that ffi_prep_cif
162 came across is bad.
163
164
165Before making the call, the VALUES vector should be initialized
166with pointers to the appropriate argument values.
167
168To call the the function using the initialized ffi_cif, use the
169ffi_call function:
170
171void ffi_call(ffi_cif *cif, void *fn, void *rvalue, void **avalues);
172
173 CIF is a pointer to the ffi_cif initialized specifically
174 for this function.
175
176 FN is a pointer to the function you want to call.
177
178 RVALUE is a pointer to a chunk of memory that is to hold the
179 result of the function call. Currently, it must be
180 at least one word in size (except for the n32 version
181 under Irix 6.x, which must be a pointer to an 8 byte
182 aligned value (a long long). It must also be at least
183 word aligned (depending on the return type, and the
184 system's alignment requirements). If RTYPE is
185 &ffi_type_void, this is ignored. If RVALUE is NULL,
186 the return value is discarded.
187
188 AVALUES is a vector of void* that point to the memory locations
189 holding the argument values for a call.
190 If NARGS is 0, this is ignored.
191
192
193If you are expecting a return value from FN it will have been stored
194at RVALUE.
195
196
197
198 An Example
199 ----------
200
201Here is a trivial example that calls puts() a few times.
202
203 #include <stdio.h>
204 #include <ffi.h>
205
206 int main()
207 {
208 ffi_cif cif;
209 ffi_type *args[1];
210 void *values[1];
211 char *s;
212 int rc;
213
214 /* Initialize the argument info vectors */
215 args[0] = &ffi_type_uint;
216 values[0] = &s;
217
218 /* Initialize the cif */
219 if (ffi_prep_cif(&cif, FFI_DEFAULT_ABI, 1,
220 &ffi_type_uint, args) == FFI_OK)
221 {
222 s = "Hello World!";
223 ffi_call(&cif, puts, &rc, values);
224 /* rc now holds the result of the call to puts */
225
226 /* values holds a pointer to the function's arg, so to
227 call puts() again all we need to do is change the
228 value of s */
229 s = "This is cool!";
230 ffi_call(&cif, puts, &rc, values);
231 }
232
233 return 0;
234 }
235
236
237
238 Aggregate Types
239 ---------------
240
241Although libffi has no special support for unions or bit-fields, it is
242perfectly happy passing structures back and forth. You must first
243describe the structure to libffi by creating a new ffi_type object
244for it. Here is the definition of ffi_type:
245
246 typedef struct _ffi_type
247 {
248 unsigned size;
249 short alignment;
250 short type;
251 struct _ffi_type **elements;
252 } ffi_type;
253
254All structures must have type set to FFI_TYPE_STRUCT. You may set
255size and alignment to 0. These will be calculated and reset to the
256appropriate values by ffi_prep_cif().
257
258elements is a NULL terminated array of pointers to ffi_type objects
259that describe the type of the structure elements. These may, in turn,
260be structure elements.
261
262The following example initializes a ffi_type object representing the
263tm struct from Linux's time.h:
264
265 struct tm {
266 int tm_sec;
267 int tm_min;
268 int tm_hour;
269 int tm_mday;
270 int tm_mon;
271 int tm_year;
272 int tm_wday;
273 int tm_yday;
274 int tm_isdst;
275 /* Those are for future use. */
276 long int __tm_gmtoff__;
277 __const char *__tm_zone__;
278 };
279
280 {
281 ffi_type tm_type;
282 ffi_type *tm_type_elements[12];
283 int i;
284
285 tm_type.size = tm_type.alignment = 0;
286 tm_type.elements = &tm_type_elements;
287
288 for (i = 0; i < 9; i++)
289 tm_type_elements[i] = &ffi_type_sint;
290
291 tm_type_elements[9] = &ffi_type_slong;
292 tm_type_elements[10] = &ffi_type_pointer;
293 tm_type_elements[11] = NULL;
294
295 /* tm_type can now be used to represent tm argument types and
296 return types for ffi_prep_cif() */
297 }
298
299
300
301Platform Specific Notes
302=======================
303
304 Intel x86
305 ---------
306
307There are no known problems with the x86 port.
308
309 Sun SPARC - SunOS 4.1.3 & Solaris 2.x
310 -------------------------------------
311
312You must use GNU Make to build libffi on Sun platforms.
313
314 MIPS - Irix 5.3 & 6.x
315 ---------------------
316
317Irix 6.2 and better supports three different calling conventions: o32,
318n32 and n64. Currently, libffi only supports both o32 and n32 under
319Irix 6.x, but only o32 under Irix 5.3. Libffi will automatically be
320configured for whichever calling convention it was built for.
321
322By default, the configure script will try to build libffi with the GNU
323development tools. To build libffi with the SGI development tools, set
324the environment variable CC to either "cc -32" or "cc -n32" before
325running configure under Irix 6.x (depending on whether you want an o32
326or n32 library), or just "cc" for Irix 5.3.
327
328With the n32 calling convention, when returning structures smaller
329than 16 bytes, be sure to provide an RVALUE that is 8 byte aligned.
330Here's one way of forcing this:
331
332 double struct_storage[2];
333 my_small_struct *s = (my_small_struct *) struct_storage;
334 /* Use s for RVALUE */
335
336If you don't do this you are liable to get spurious bus errors.
337
338"long long" values are not supported yet.
339
340You must use GNU Make to build libffi on SGI platforms.
341
342 ARM - System V ABI
343 ------------------
344
345The ARM port was performed on a NetWinder running ARM Linux ELF
346(2.0.31) and gcc 2.8.1.
347
348
349
350 PowerPC System V ABI
351 --------------------
352
353There are two `System V ABI's which libffi implements for PowerPC.
354They differ only in how small structures are returned from functions.
355
356In the FFI_SYSV version, structures that are 8 bytes or smaller are
357returned in registers. This is what GCC does when it is configured
358for solaris, and is what the System V ABI I have (dated September
3591995) says.
360
361In the FFI_GCC_SYSV version, all structures are returned the same way:
362by passing a pointer as the first argument to the function. This is
363what GCC does when it is configured for linux or a generic sysv
364target.
365
366EGCS 1.0.1 (and probably other versions of EGCS/GCC) also has a
367inconsistency with the SysV ABI: When a procedure is called with many
368floating-point arguments, some of them get put on the stack. They are
369all supposed to be stored in double-precision format, even if they are
370only single-precision, but EGCS stores single-precision arguments as
371single-precision anyway. This causes one test to fail (the `many
372arguments' test).
373
374
375What's With The Crazy Comments?
376===============================
377
378You might notice a number of cryptic comments in the code, delimited
379by /*@ and @*/. These are annotations read by the program LCLint, a
380tool for statically checking C programs. You can read all about it at
381<http://larch-www.lcs.mit.edu:8001/larch/lclint/index.html>.
382
383
384History
385=======
386
3871.20 Oct-5-98
388 Raffaele Sena produces ARM port.
389
3901.19 Oct-5-98
391 Fixed x86 long double and long long return support.
392 m68k bug fixes from Andreas Schwab.
393 Patch for DU assembler compatibility for the Alpha from Richard
394 Henderson.
395
3961.18 Apr-17-98
397 Bug fixes and MIPS configuration changes.
398
3991.17 Feb-24-98
400 Bug fixes and m68k port from Andreas Schwab. PowerPC port from
401 Geoffrey Keating. Various bug x86, Sparc and MIPS bug fixes.
402
4031.16 Feb-11-98
404 Richard Henderson produces Alpha port.
405
4061.15 Dec-4-97
407 Fixed an n32 ABI bug. New libtool, auto* support.
408
4091.14 May-13-97
410 libtool is now used to generate shared and static libraries.
411 Fixed a minor portability problem reported by Russ McManus
412 <mcmanr@eq.gs.com>.
413
4141.13 Dec-2-96
415 Added --enable-purify-safety to keep Purify from complaining
416 about certain low level code.
417 Sparc fix for calling functions with < 6 args.
418 Linux x86 a.out fix.
419
4201.12 Nov-22-96
421 Added missing ffi_type_void, needed for supporting void return
422 types. Fixed test case for non MIPS machines. Cygnus Support
423 is now Cygnus Solutions.
424
4251.11 Oct-30-96
426 Added notes about GNU make.
427
4281.10 Oct-29-96
429 Added configuration fix for non GNU compilers.
430
4311.09 Oct-29-96
432 Added --enable-debug configure switch. Clean-ups based on LCLint
433 feedback. ffi_mips.h is always installed. Many configuration
434 fixes. Fixed ffitest.c for sparc builds.
435
4361.08 Oct-15-96
437 Fixed n32 problem. Many clean-ups.
438
4391.07 Oct-14-96
440 Gordon Irlam rewrites v8.S again. Bug fixes.
441
4421.06 Oct-14-96
443 Gordon Irlam improved the sparc port.
444
4451.05 Oct-14-96
446 Interface changes based on feedback.
447
4481.04 Oct-11-96
449 Sparc port complete (modulo struct passing bug).
450
4511.03 Oct-10-96
452 Passing struct args, and returning struct values works for
453 all architectures/calling conventions. Expanded tests.
454
4551.02 Oct-9-96
456 Added SGI n32 support. Fixed bugs in both o32 and Linux support.
457 Added "make test".
458
4591.01 Oct-8-96
460 Fixed float passing bug in mips version. Restructured some
461 of the code. Builds cleanly with SGI tools.
462
4631.00 Oct-7-96
464 First release. No public announcement.
465
466
467Authors & Credits
468=================
469
470libffi was written by Anthony Green <green@cygnus.com>.
471
472Portions of libffi were derived from Gianni Mariani's free gencall
473library for Silicon Graphics machines.
474
475The closure mechanism was designed and implemented by Kresten Krab
476Thorup.
477
478The Sparc port was derived from code contributed by the fine folks at
479Visible Decisions Inc <http://www.vdi.com>. Further enhancements were
480made by Gordon Irlam at Cygnus Solutions <http://www.cygnus.com>.
481
482The Alpha port was written by Richard Henderson at Cygnus Solutions.
483
484Andreas Schwab ported libffi to m68k Linux and provided a number of
485bug fixes.
486
487Geoffrey Keating ported libffi to the PowerPC.
488
489Raffaele Sena ported libffi to the ARM.
490
491Jesper Skov and Andrew Haley both did more than their fair share of
492stepping through the code and tracking down bugs.
493
494Thanks also to Tom Tromey for bug fixes and configuration help.
495
496Thanks to Jim Blandy, who provided some useful feedback on the libffi
497interface.
498
499If you have a problem, or have found a bug, please send a note to
500green@cygnus.com.
501